Is Your Business AI-Ready? A Practical Checklist for East African Companies

"We want to use AI" is usually the wrong starting point — it's a solution looking for a problem. The businesses that get real value from AI almost always start somewhere more boring: do we actually have the data this would need, is it in a state anyone could use, and is there a specific decision this would improve? Skip that groundwork and you end up with an expensive pilot that never makes it past a slide deck. Here's the checklist we walk clients through before we agree to build anything.

1. Do you actually have the data this would need?

Not "big data" in the abstract — the specific data relevant to the decision you're trying to improve, collected consistently enough over time to be useful. A business can have terabytes of logs and still not have the one field that actually matters, recorded reliably, for long enough to train anything on.

2. Is that data in a state anyone could actually use?

Scattered spreadsheets across departments, inconsistent formats, no single source of truth for a customer or transaction record — these problems don't go away because you bought an AI tool. Data engineering usually has to happen before modeling, not after, and skipping it is the single most common reason a promising pilot never reaches production.

3. Is there a specific decision this would improve?

"We want to use AI" isn't a starting point — it's a symptom of skipping this step. "We want to cut manual document review time by half" or "we want to flag high-risk transactions before they clear" is a starting point, because it gives you something concrete to measure against.

4. Does someone inside the business actually own this?

AI initiatives that float without an internal sponsor accountable for the outcome tend to stall at the pilot stage regardless of how good the model is. Someone needs to own adoption, own the decision to change a process because of what the model says, and own the uncomfortable conversation if it doesn't work the first time.

5. Can the business tolerate being wrong sometimes?

Every model is probabilistic. If a wrong prediction in your use case is catastrophic and unrecoverable, you need a human-in-the-loop process and a clear fallback, not a fully automated decision from day one. Knowing this in advance changes how you design the system, and it's far cheaper to design for it upfront than to retrofit it after an incident.

What happens when you skip this

The common failure pattern looks the same across sectors: a proof of concept impresses everyone in a demo, then quietly dies in the months after because nobody defined what "working" meant, nobody owned turning it into a real process change, and the underlying data problems that were papered over for the demo resurface at scale.

How we approach this with clients

This is why an AI-readiness audit is usually the first engagement we recommend, not the model itself — a short, honest diagnostic against these five questions, before anyone commits budget to building anything. It's a cheaper conversation to have upfront than after a pilot has already quietly failed.